I want to predict which proteins or DNA or RNA might interact with a given protein. However mostly whatever I find is that databases like StringDB etc actually list the already known interacting partners based on pubmed text mining. One trivial way to go is that make a blast search with already known interacting protein or specific domain in order to find out something similar partners or to look at the domains the protein contain and with which proteins these domains interact. However this mostly gives results whch are already known. Is there any more sophistaced way of doing it by machine learning or evolutionary algorithms? Or I should rely on PSI-BLAST for distant homologies?
I think that PSI-BLAST might help predict if your protein of interest is likely to be able to bind DNA or RNA, but predicting specific target sequences is more difficult. It depends in part on the family to which your protein belongs e.g. the target sequences bound by some zinc-finger proteins can be predicted. As to protein-protein interaction (PPI) partner prediction, then one method might be that used by Pitre and colleagues or Liu et al - see: http://www.nature.com/srep/2012/120130/srep00239/full/srep00239.html or http://www.plosone.org/article/info:doi/10.1371/journal.pone.0030938?utm_source=feedburner They both use cross-species conservation, but in different ways.